The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill instructs users to install global packages btca and opencode-ai from unverified third-party sources using bun add -g. This introduces supply chain risks.
[REMOTE_CODE_EXECUTION] (MEDIUM): The instructions suggest downloading an instruction file from https://btca.dev/rule directly into .cursor/rules/. While formatted as markdown, these .mdc files define behavior for AI agents, allowing remote actors to influence or override agent logic.
[DATA_EXFILTRATION] (LOW): The tool supports a btca serve command which exposes indexed repository data over an HTTP server on port 3000. If misconfigured or exposed to the network, this could lead to unauthorized access to local source code.
[PROMPT_INJECTION] (LOW): The skill is designed to ingest and query untrusted content from external Git repositories (Category 8). This creates a surface for indirect prompt injection where malicious instructions embedded in a repository could manipulate the agent's output.
Ingestion points: Git repositories cloned to ~/.local/share/btca/resources/ (e.g., https://github.com/bknd-io/bknd).
Boundary markers: None specified in the instructions; protection depends on the internal implementation of the btca tool.
Capability inventory: Terminal command execution, network access for repository cloning, and local server hosting.
Sanitization: Not explicitly mentioned; the skill relies on the underlying LLM's safety filters and the btca tool's handling of content.
[COMMAND_EXECUTION] (LOW): The skill provides numerous CLI commands for repository indexing and querying that are intended for manual execution by the user.

btca-bknd-repo-learn